The Review of Socionetwork Strategies Бизнес и менеджмент

Enhanced Skin Cancer Detection Using Deep and Handcrafted Feature Fusion with Ensemble Classification

Bhargavi, Maridu, Balakrishna, Sivadi
Springer Nature · 2026-10-01 · DOI 10.1007/s12626-026-00219-8

Аннотация

The early and accurate identification of skin cancer is critical for improving patient survival rates and minimizing the overall burden on the healthcare system. This study proposes a hybrid feature fusion framework for multi-class skin cancer classification that integrates deep feature learning from a convolutional neural network (CNN) with handcrafted features extracted through ensemble methods for final classification, resulting in a highly reliable and scalable approach to intelligent skin cancer diagnosis. The workflow consists of preprocessing dermoscopic images from the PAD-UFES-20 and ISIC datasets, extracting deep features through a custom deep convolutional neural network (DCNN) architecture, and extracting complementary handcrafted features, including Gray-Level Co-occurrence Matrix (GLCM) texture descriptors, color descriptors, and Completed Local Binary Pattern (CLBP) descriptors, before combining these features into a unified representation. The fused feature set is then subjected to dimensionality reduction using Principal Component Analysis (PCA) before final classification, which is performed using an ensemble of Histogram Gradient Boosting (HGB), Extreme Gradient Boosting (XGBoost), and Random Forest (RF) classifiers combined through a soft voting approach. The experimental evaluation demonstrated that the proposed framework provides a robust, accurate, and scalable approach for computer-aided skin cancer diagnosis. The model achieved classification accuracies of 95.17% on the PAD-UFES-20 dataset and 93.86% on the ISIC dataset, while consistently attaining approximately 95%, 94%, and 94% for precision, recall, and F1-score, respectively. By leveraging complementary feature representations and ensemble learning methods, the proposed approach achieved high levels of accuracy, reliability, and generalizability.

Читать у издателя →